AI can help transform our future for the better
Transforming biological sciences with AI will require innovation both in the experimental as well as computational technologies, and will encompass revisiting and challenging multiple aspects currently done in a traditional way: from the design and execution of biological and chemical experiments in a wet lab, data collection and analysis, to simulation of molecules, cells, tissues and organs, and drug discovery and design.
A longer-term impact will be in computational science (new machine learning architectures and methods), fundamental science (furthering our understanding of biological systems and how life works), and benefitting the humankind (better understanding disease and developing better, faster, and more reliable diagnostics and cures to currently incurable diseases).
Developing AI tools to foster biological sciences
We believe that existing similar initiatives are likely to be synergistic with AITHYRA and we see them as potential collaborators rather than competitors. To the maximum extent possible, AITHYRA will strive to develop AI tools that can be used to address problems at various biological scales. AITHYRA will drive impact in the 5-10 years in the following areas:
Develop…
…a “successor to Transformers,” a class of scalable and efficient deep neural network architectures suitable for a broad range of AI problems in biological sciences.
…AI-based molecular simulators that would allow to accelerate orders of magnitude traditional tools based on numerical differential equations, and use them to create and release accurate simulated datasets that can be used to train higher-level and task-specific AI models.
…portable “foundational biological AI models” that would bridge multiple scales and could be adapted across a range of applications.
…generative AI methods for molecular generation that can be used alongside simulators to generate large-scale synthetic datasets, to answer fundamental biological questions, as well as for commercial applications such as drug discovery.
…and promote, with industrial stakeholders, safe and secure methodologies for training AI models on sensitive and proprietary data. Incorporate these methodologies into foundational AI models and train them on data pooled from a consortium of pharmaceutical companies.
Contribute…
…to existing open-source software libraries and introduce our own.
….to existing open-source datasets and benchmarks and introduce our own.
Translate…
…scientific results into the market or clinical practice, e.g. through industrial collaborations, technology licensing, spin-offs, and clinical trials.